PewDiePie has unveiled Ajax, an "uncensored AI model" designed to run at home as part of his self-hosted Odysseus workspace. The model, a fine-tuned Qwen3.5-9B, is meant to handle daily tasks such as web search, email, and calendar management entirely privately. According to its project page, Ajax's refusal behavior has been removed for a "freer, less restricted AI experience," though users are asked to use it responsibly.
In the announcement video, PewDiePie says OpenAI banned him twice while he was developing the model. One ban was explicitly for "distillation," the practice of using another model's outputs or reasoning to train a new model. He admitted he wanted to "distill just a little bit" to improve his model, and questioned how OpenAI detected the second run. The article notes that frontier labs like OpenAI and Anthropic have been working to hide their reasoning and treat attempted distillation as a threat.
To remove refusals, PewDiePie relied on automatic abliteration through the open-source Heretic tool, which identifies refusal directions and removes them without, ideally, causing "brain damage" to the model. He says he draws the line at content that harms other people or oneself, and on his lawyer's advice, Ajax is "not designed to provide dangerous actionable instructions." The exact limits remain unclear.
PewDiePie says he strongly believes in small models trained for specific harnesses rather than trillion-parameter frontier models, and he asked fans to donate training data as an alternative to collecting it from Odysseus users, though none submitted any. He plans two more days of reinforcement learning before rerunning the decensoring step, quantizing, and benchmarking, with a next version of Ajax and Odysseus promised. The article frames the project as part of a broader movement toward less restricted intelligence on users' own hardware.